In the current Healthcare climate, managed care plans provide for an increasing proportion of health services to populations from fixed pools of dollars. In capitated health plans, physicians are reimbursed based on the anticipated healthcare needs of the population they serve. Therefore, fair and equitable distribution of healthcare dollars between health plans and among physicians within plans requires an accurate description of the health conditions and service needs of the enrolled population. There are a number of different classification systems with the population group as the basis of analysis, each with its own strengths and limitations. This article briefly reviews these systems, then describes and illustrates the uses of the Classification of Congenital and Chronic Health Conditions, a diagnosis-based classification system recently developed by the National Association of Children's Hospitals and Related Institutions (NACHRI).
POPULATION-BASED CLASSIFICATION METHODS
Classification systems used to describe populations are distinguished from those in which the unit of analysis is an encounter or episode of care (such as diagnosis-related groups used to classify hospital discharges). Examples include the following:
1 . Age and gender are demographic variables commonly used to classify populations, and are readily available and resistant to manipulation. Although the data provide some information about potential health service needs, (eg, immunization services to young children, obstetrical and gynecologic services to women of childbearing age), the very general information provided does not differentiate between sickness and wellness within categories. The limited power for predicting current or future health service utilization and cost is well-documented. l"3
2. Prior health service utilization models have greater predictive power than age and gender, but provide neither an understanding of population groups nor incentives for cost management. These models do not necessarily relate to the future course of an illness.4"6
3. Health status questionnaires, which generally ascertain an individual's self-perceived health status, can provide useful information with regard to functional status, but require administration and regular updating, are subjective and subject to manipulation, and are not powerful predictors of health service utilization and costs.7
4. Diagnosis-based classification models can identify specific diseases and disease combinations that are predictive of health service utilization and costs and are administratively feasible if a claims or encounter-level database is available. Their limitations include insufficient specificity of some diagnosis codes and variability in the clinical course of many diseases. The NACHRI Classification of Congenital and Chronic Health Conditions (CCCHC) is such a system.
DESIGN AND STRUCTURE OF NACHRI CLASSlHCATION OF CONGENITAL AND CHRONIC HEALTH CONDITIONS
The NACHRI system provides a conceptual and operational means through International Classification of Diseases, 9th Edition-Clinical Modification (ICD-9-CM) codes to identify individuals with a chronic health condition. The system also classifies certain at-risk categories to the extent possible with existing diagnostic codes. The addition of age and gender completes the description from which to derive estimates of expected health service needs for each population. Acute illnesses are tracked only to further describe the status of a chronic condition. Acute illness and health maintenance care, although representing a large component of the population's health service needs, are generally not very predictive of future health service needs.2
The CCCHC is a clinical model designed to develop as complete a clinical profile of the population as practical from readily available information on the diagnostic conditions of enrolled individuals. Its development began with the identification of clinically meaningful and recognizable groupings of patients, accomplished with a wealth of input from practicing general and subspecialty pediatricians and surgeons. Next, the ICD-9-CM diagnosis codes that represent chronic health conditions were identified. Then, the case volume of the chronic disease diagnoses and statistics on health service utilization and costs were identified using databases from the State of Washington Medicaid program and a private health insurance dataset.
The following are concepts and definitions key to understanding the structure, capabilities, and limitations of the NACHRI classification system:
Chronic Health Condition: Physical, mental, emotional, behavioral, or developmental disorder expected to last 12 months or longer or having sequelae that last 12 months OT longer and requiring treatment and/or monitoring. Approximately 4000 of the 15 000 current ICD-9-CM diagnosis codes meet this criterion.
Major Diagnostic Category (MDCJ : Major body systems (eg, cardiovascular and musculoske letal) and major disease processes (eg, blood disorders, malignant neoplasms).
Individual Chronic Condition Category: Clinically distinct, recognizable categories, defined on the basis of case volume, anatomy or physiology, disease progression, congenital or inherited versus acquired, and interaction between conditions. There are approximately 250 chronic condition categories, of which 40 are mental health categories.
Supplementul Status indicators: Supplemental status V codes and related ICD-9-CM diagnostic codes that identify major medical assistive devices and ongoing treatment modalities, providing additional information about the status of persons with chronic health conditions. For example, chemotherapy status indicates that a malignant neoplasm is in the active treatment phase.
Severity Level of individual Diagnoses: Each diagnosis is given one of four initial severity level assignments - mild (Sl), moderate (S2), major (S3), and extreme (S4) - based on expected complexity and costliness of all healthcare services over a 12-month period.
Severity Level of Individuals: Severity-level assignments for each person take into account the severity level of individual diagnoses, disease progression, interactive effects of multiple conditions, and supplemental status indicators. For example, congenital anomalies that can be cured or substantially improved may receive high severity-level assignments in die first year of life followed by lower severity levels thereafter.
Disease Progression: Each diagnosis is assigned to one of the following disease progression types based on the expected disease course and treatment goal: cure/substantially improve, substantially improve/continuous treatment, static/improve function, progressive, supportive care, or mixed course.
"At-Rislc" Categories: Conditions that do not meet the criteria used to define chronic conditions fie, expected to last or have sequelae that last 12 months or longer at least 75% of the time), but usually require services of an amount and type greater than that for non-chronically ill persons and place the individual at risk for ongoing chronic conditions. Examples include prematurity, prenatal drug or alcohol exposure, failure to thrive, and child abuse or neglect. Psychological diagnoses, such as adjustment disorder with mood or conduct disturbance, have been included also and are undergoing further review.
The NACHRI classification system can be used to assist managed care organizations and physician groups to better understand and serve their many populations by:
1. Identifying health service needs through establishment of disease prevalence rates for the full spectrum of chronic physical and mental health conditions and certain at-risk categories.
2. Organizing the delivery of healthcare services based on frequency and types of chronic health conditions as well as the age and gender of the population served.
3. Tracking quality indicators and patient and family satisfaction for distinctive subgroups of the population, such as the chronically ill.
4. Profiling the practice patterns of individual physicians, physician groups, and entire health plans, case-mix-adjusted for the prevalence and types of chronic health conditions.
5. Risk-adjusting capitation payment rates based on the prevalence and types of chronic health conditions.
The limitations of the NACHRI system include the following:
1. The ICD-9-CM codes do not always provide the degree of specificity necessary to fully describe the condition of an individual patient.
2. With regard to provider profiling, physicians who, by virtue of specialization, treat large numbers of acute conditions in non-chronically ill persons will not have this reflected in case-mix measurements by the NACHRI classification system.
USES OF NACHRl CLASSIFICATION OF CONGENITAL AND CHRONIC HEALTH CONDITIONS
The uses of the NACHRI classification system will be illustrated from its first testing on a fullservice paid claims database from the State of Washington Medicaid program. The State of Washington Medicaid database includes non-institutionalized recipients, aged O to 64 years, with eligibility for service for at least 9 months of fiscal years 1992 and 1993. Total billed charges for each recipient were annualized based on the number of months the recipient was eligible for services. There were 417 808 recipients meeting this eligibility requirement and passing "edit" tests, the most important of which was removing mothers whose paid claims records included the diagnoses and bills for their newborn children. Of the total recipients, 271318 (64.9%) were children (ages O to 17 years), including 7735 SSI disabled recipients; 146 490 (35.1%) were adults (ages 18 to 64 years), including 48 285 SSI disabled recipients. Among all recipients, 95.9% had one or more paid claims for health services in this time period.
The database and classification system used for this analysis differ from those in a previous report8 in two respects. First, the current database included claims experience for recipients for up to 24 months rather than 12 months, increasing the likelihood of identifying a chronic disease diagnosis. Second, an updated version of the classification system was used, which included more chronic disease diagnoses and took into account the coding of diagnoses in outpatient settings, with diagnoses often coded less completely yet sufficiently to identify the presence of a chronic health condition.
For this analysis, chronic disease prevalence and cost profiling were performed for children and adults in the Medicaid population at the following levels of aggregation: presence or absence of a chronic condition, severity level of the condition, major diagnostic category (body system), individual chronic condition category, and presence of multiple chronic conditions.
As shown in Table 1, the overall chronic disease prevalence rate for children was 23.2%, with rates slightly higher for infants and adolescents (25.3%) and slightly lower for preschool- and grade school-aged children (21.8%). The overall rate for adults was 54.8%.
Table 2 presents chronic disease prevalence rates, total billed charges, and physician-billed charges by severity level. Individuals with the higher severity levels represented a small percentage of the population but a significant percentage of total billed charges for healthcare services. This relationship also applied to individuals with moderate and mild chronic conditions, although less dramatically. In the pediatrie Medicaid population, the 23.2% with chronic illnesses accounted for 61.8% of all billed charges for health services. The average annual total billed charges for children without a chronic condition was $812, whereas that for children with a chronic condition was $4350 (5.36-fold difference). For adults, there was a very similar 4.89-fold increase in charges for individuals with chronic conditions identified by the NACHRI classification system.
Overall Chronic Illness Prevalence Rates*
Chronic Disease Prevalence Rates and Charges By Severity Level*†
In the Washington State Medicaid dataset, billing information is broken down into service categories. As a result, physician billing can be analyzed separately. Table 2 also shows physician billing for individuals with chronic conditions. In general, these billing data followed the same pattern as total billed charges, children with chronic conditions having billings 3.27-fold higher than those without ($916 versus $280 per annum). Physician charges increased at a percentage similar to total billed charges for patients with mild and moderate chronic conditions. Physician charges continued to increase substantially for patients with major and extreme chronic illnesses, although not quite as dramatically as total billed charges. Presumably, this reflects the exceptional costs for hospital care, home health care, and other very expensive interventions for these children.
Although not shown in Table 2, the Washington State Medicaid dataset also contains amounts billed and paid for many different services, which can be tracked with the NACHRl system. Thus, it is possible to examine the proportion of billed or paid dollars applied to hospital care, physician services, drugs, durable medical equipment, or other service categories for various groups of children.
Table 3 presents chronic disease prevalence rates by body system and Table 4 by individual chronic condition category for the most common chronic illnesses. Because these are counts of conditions, individuals with multiple chronic conditions will be represented multiple times. For children, respiratory conditions were most prevalent at 7.3%, followed by mental health (6.2%), musculoskeletal (3.0%), and nervous system disorders (2.8%). For adults, chronic mental health conditions as a group were most prevalent. As shown in Table 4, asthma was the most common respiratory condition in children (as in adults), whereas the most common nervous system condition was epilepsy. In the endocrine system, diabetes mellitus was by far the most common condition, with thyroid gland disorders common for both adults and children.
Chronic Illness Prevalence By Body System*†
Chronic Illness Prevalence By Condition Category*†
There were dramatic differences in die prevalence of specific conditions among the SSI-disabled and nonSSI-disabled populations (data not shown). For example, there was a much higher percentage of children with mental retardation, mental illnesses, and certain neurologic conditions such as cerebral palsy, spina bifida, and muscular dystrophy among the SSI disabled.
Table 5 presents average per annum total billed charges by age, gender, and chronic condition. The data demonstrate that the presence of diagnoses identifying chronic conditions and "at-risk" conditions had far more impact than age and gender for predicting health service utilization and costs. Of those with chronic health conditions, most of the differences in average charges per age grouping were attributable to chronic disease severity levels and the presence of multiple chronic conditions (Figures 1 and 2). Of those with at-risk conditions, charges were similar for various age and gender categories except that newborns and adolescents were slightly more expensive. The most frequent at-risk conditions of those studied were failure to thrive (n=5939), adjustment disorder (n=4124), and child abuse or neglect (n=1058). Of those with no chronic condition, charges were similar for various age and gender categories except that newborns and women of childbearing age were somewhat more expensive and children 5 to 10 years of age were least expensive.
Average per annum charges by chronic disease severity level for children and adults are shown in Figure 1 . It shows a very similar pattern for adults and children with average charges increasing two- to threefold from the non-chronically ill to those with mild chronic conditions and likewise from one chronic condition severity level to the next. Average charge levels for adults and children with major (S3) and extreme (S4) chronic conditions were nearly identical. Average charge levels were a little higher for adults than children with moderate (S2) and mild (Sl) chronic conditions. This relates primarily to the presence of multiple chronic conditions in adults (data not shown). The non'chronicaUy ill adults were also a little more expensive than children, likely attributable to childbirth expenses for women. Notwithstanding, the overall pattern for this first-cut viewing of charge levels showed great similarity between adults and children. Not shown in Figure 1 is the average charge levels among different adult age groups. These charge levels were especially consistent, whereas average charge levels were more variable among pediatrie patients, with newboms, infants or toddlers, and adolescents being somewhat more expensive than children ages 2 to 4 years and 5 to 10 years.
Average Per Annum Billed Charges By Age, Gender, and Chronic Diagnosis*†
Figure 1. Per annum charges by severity level for children and adults. State of Washington Medicakl, fiscal years 1 992 and 1993. Non-institutionalized recipients, aged O to 64 years.
Figure 2. Per annum charges by severity level and combinations for all Medicaid patients. State of Washington Medicaid, fiscal years 1992 and 1993. Non-institutionalized recipients, aged O to 64 years.
Figure 3. Per annum charges by presence of chronic condition and SSI disabled status. State of Washington Medicaid, fiscal years 1992 and 1993. Non-institutionalized recipients, aged O to 64 years.
Figure 2 presents average per annum charges by severity level and presence of multiple chronic conditions. This adds greatly to the predictive power of the system. To illustrate, children with one or more moderate chronic conditions had average annual charges of $4350. Those with multiple moderate chronic conditions from different body systems had average charges of $9840; those with multiple chronic conditions from the same body system had average charges of $8272; and those with a single chronic condition had average charges of $3359. Thus, within the range of moderate chronic conditions, there was a threefold difference in charges based on the presence of multiple conditions and involvement of different body systems.
The final version of the software for the NACHRI classification system places all persons into mutually exclusive categories. This will provide greater differentiation in average cost levels, less variability within categories, and greater clinical relevance, enabling more effective provider profiling and capitation risk adjustment.
Our analysis also uncovered marked differences in average per annum charges by presence or absence of a chronic health condition for SSI disabled and nonSSI disabled Medicaid recipients (Figure 3 ) . The presence of a chronic health condition was a powerful predictor, because those who did not have chronic health conditions had consistently lower charge levels. The charge levels were particularly high for the SSI disabled with chronic health conditions, reflecting a greater proportion of high-severity chronic diseases and a greater proportion of multiple conditions.
It is particularly noteworthy that the SSI disabled recipients lacking a chronic disease diagnosis identified by the classification system had billed charges for healthcare services that were essentially the same as those for the non-SSI disabled without a chronic disease diagnosis. These individuals may have had a chronic disabling condition which had resolved but for whom recertification had not occurred, or still had a chronic disabling condition but were not seeking or receiving healthcare services. From a classification perspective, the significance lies in the ability to identify such persons and situations. Once identified, focused management review can occur, allowing a greater understanding of the population's health needs and their fulfillment.
The utility of any medical classification system lies in its clinical credibility, which depends on the concepts, definitions, and information building blocks used in its creation. These components were considered very rigorously in the design and testing of the NACHRI classification system. The system was developed to be comprehensive of chronic conditions for patients of all ages. The first testing suggests that the statistical performance may be as strong for adults as for children, although the system has not had the benefit of extensive review by internists and non-pediatric surgeons. This is intended for the next version of the system.
The informational building blocks for the NACHRI classification system are the ICD-9-CM diagnosis codes and supplemental status codes. Although they are often criticized for lacking sufficient specificity, there is a wealth of information in the 15 000 ICD-9-CM diagnosis codes. There are also established coding principles and guidelines, and the codes are revised annually. Furthermore, ICD-IO, currently in use in Europe and Australia and slated for use in the United States beginning in 2001, contains a number of improvements over ICD-9-CM and will be incorporated into the NACHRI system.
To realize the full potential of the NACHRI system, it is important to (1) understand and make maximum use of the existing information in the ICD-9-CM codes; (2) encourage and support the full and accurate reporting of all pertinent ICD-9-CM diagnosis and supplemental status codes by all physicians and all health plans for all patient care encounters, particularly in the outpatient setting; and (3) continue to formulate proposals to the ICD9-CM Coordination and Maintenance Committee for new and revised codes.
Even with complete and accurate recording of ICD-9-CM diagnosis codes, there is not always sufficient information for a full description of a person's medical condition and required medical services. Furthermore, predicting the course of an illness and necessary interventions is not always possible. The NACHRI strategy is to identify the most essential information that is readily and practically available to describe a person's health conditions and needs.
Some conditions lend themselves to further characterization by supplemental information on stage of illness or functional status. This information also can be inferred from viewing all of a patient's diagnoses and the interactive effects of the disorders, which is the approach of the NACHRI classification system. For example, a person with diabetes mellitus who also has peripheral neuropathy or nephropathy has more advanced disease. Supplemental status V codes can also be useful, eg, wheelchair-dependent status, recently considered as a new code, would predict a worsening clinical course in a patient with muscular dystrophy.
An important practical issue is the frequent reliance on diagnostic information from billing and encounter forms used in the physician office setting. In many practices and institutions, the encounter form focuses on procedure codes for billing purposes and diagnostic information is streamlined, providing only the most general type of medical condition reporting and limiting the degree of detail available as to the condition of an individual patient. However, the encounter form provides an opportunity to broaden the diagnostic information coded. If incorporated into standard billing forms, more complete diagnostic coding could become more automatic, resulting in a more accurate clinical representation of the patient.
In addition to its uses for managed care, the NACHRI classification system has public health and policy applications. For example, identification and tracking for children with special health needs is feasible. This would require that health plans maintain datasets and contribute to a central data repository on a regular basis. Once created, a central data repository would allow for tracking the chronically ill over time as they enroll, disenroll, and re-enroll in different health plans.
The NACHRI classification system provides a tool to identify the population of "children with special healthcare needs." This term is widely used but poorly defined. In some circumstances, the term is used to describe all chronically ill children plus an "at-risk" population and in others, it describes those with the most complex conditions.9,10 These differences often reflect the use of the term in different contexts and for different purposes. An operational definition can be developed only if the purpose and intended scope is clear. Once established, an operational definition could be developed working with the categories and severity levels of the NACHRI classification system.
The identification of health plan enrollees with higher than average healthcare costs, such as those with chronic health conditions, has significant implications for physician profiling and payment. From the analysis of the Washington State Medicaid data, it is important to recognize that children are not necessarily less expensive to treat than adults; rather, the presence and type of chronic condition is a more powerful determinant of needs than age and gender alone. Although children in general are healthier and have less chronic disease than adults, physicians and health plans that provide care for a greater number of children who do have chronic conditions must be capitated appropriately for the population under their care. This is especially important because, unlike adults, relatively few children have chronic conditions and much of their care is provided by a small number of physicians and hospitals specializing in their treatment. The NACHRI system provides a means for accurate identification of this population.
The authors would like to acknowledge the contributions of the State of Washington for authorizing the use of its Medicaid database and the Johns Hopkins University Center for Hospital Finance and Management for its programming to create personbased analytic files from the Washington Medicaid database. They would also like to acknowledge the contribution of 3M Health Information Systems for its role in programming and applying the NACHRl classification system to the Washington Medicaid database, in particular, James Vertrees, PhD, Elizabeth C. Finneran, MS, and Mona Z. Zhang, BS. Finally, they would like to acknowledge the contribution of Lisa Turner, Senior Administrative Assistant, NACHRI, for preparation of the manuscript and supporting tables.
1. Newhouse JP, Manning WG, Keeler EB, Sloss EM. Adjusting capitation rates using obiettive health measures and prior utilization. Health Care Financing Review. 1994:10:41-54.
2. Ellis R, Pope O. Ienonni LI, et al. Diagnosis-based risk adjustment for Medicare capitation payments. Health Care Financing Review. 1996; 17:101- 12T.
3. Weiner J, Dobson A, Maxwell S, Coleman K, Starfieid B, Anderson G. Risk adjusted Medicare capitation ratei using ambulatory and inpatient diagnoses. Health Care Financing Review. 1996;I7:77-99.
4. Beebe J, Lubitj J, Eggers P. Using prior utilization Information to determine payments for Medicare enrollees in HMOs. Health Care Financing Review. 1985;6:27-38.
5. Thomas JW, Lichtenscein R. Including health status in Medicare's adjusted average per capita cost capitation formula. Med Cart. 1986; 24: 2 59-2 75.
6. Kronick R, Dreyfus T, Lee L, Zhou Z- Diagnostic risk adjustment for Medicaid: the disability payment system. Heoln Care Financing Review. 19%;17:7-33.
7. Fowles JB, Lawthers A, Weinet J. Agreement between physicians' orfice records and Medicare part B clauca data. HeaIA Care Financing Review. 1995; 16: 1 89- 200.
8. Muldoon JH, Neff JM, Gay JC. Profiling the health service needs of populations using diagnosis-based classification systems. Journal of Ambulatory Care Management. 1997;20:1-18.
9. Division of Services of Children with Special Health Care Needs. Children with spedai heath care needs m managed cara organitattimi. Deftniaans and identification, family participation, capitations and rislc adjustment, quality of care. Summary of expert work group meetings. Rockville, MD; Maternal and Child Health Bureau, US Department of Health and Human Services; 1996.
10. Olympia WA. Children with special health care needs repon. Washington State Health Care Advisory Board; 1997.
Overall Chronic Illness Prevalence Rates*
Chronic Disease Prevalence Rates and Charges By Severity Level*†
Chronic Illness Prevalence By Body System*†
Chronic Illness Prevalence By Condition Category*†
Average Per Annum Billed Charges By Age, Gender, and Chronic Diagnosis*†